]> CyberLeo.Net >> Repos - FreeBSD/releng/10.2.git/blob - sys/kern/sys_process.c
- Copy stable/10@285827 to releng/10.2 in preparation for 10.2-RC1
[FreeBSD/releng/10.2.git] / sys / kern / sys_process.c
1 /*-
2  * Copyright (c) 1994, Sean Eric Fagan
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *      This product includes software developed by Sean Eric Fagan.
16  * 4. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_compat.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/syscallsubr.h>
42 #include <sys/sysent.h>
43 #include <sys/sysproto.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/vnode.h>
47 #include <sys/ptrace.h>
48 #include <sys/rwlock.h>
49 #include <sys/sx.h>
50 #include <sys/malloc.h>
51 #include <sys/signalvar.h>
52
53 #include <machine/reg.h>
54
55 #include <security/audit/audit.h>
56
57 #include <vm/vm.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_kern.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_param.h>
65
66 #ifdef COMPAT_FREEBSD32
67 #include <sys/procfs.h>
68 #include <compat/freebsd32/freebsd32_signal.h>
69
70 struct ptrace_io_desc32 {
71         int             piod_op;
72         uint32_t        piod_offs;
73         uint32_t        piod_addr;
74         uint32_t        piod_len;
75 };
76
77 struct ptrace_vm_entry32 {
78         int             pve_entry;
79         int             pve_timestamp;
80         uint32_t        pve_start;
81         uint32_t        pve_end;
82         uint32_t        pve_offset;
83         u_int           pve_prot;
84         u_int           pve_pathlen;
85         int32_t         pve_fileid;
86         u_int           pve_fsid;
87         uint32_t        pve_path;
88 };
89
90 struct ptrace_lwpinfo32 {
91         lwpid_t pl_lwpid;       /* LWP described. */
92         int     pl_event;       /* Event that stopped the LWP. */
93         int     pl_flags;       /* LWP flags. */
94         sigset_t        pl_sigmask;     /* LWP signal mask */
95         sigset_t        pl_siglist;     /* LWP pending signal */
96         struct siginfo32 pl_siginfo;    /* siginfo for signal */
97         char    pl_tdname[MAXCOMLEN + 1];       /* LWP name. */
98         int     pl_child_pid;           /* New child pid */
99 };
100
101 #endif
102
103 /*
104  * Functions implemented using PROC_ACTION():
105  *
106  * proc_read_regs(proc, regs)
107  *      Get the current user-visible register set from the process
108  *      and copy it into the regs structure (<machine/reg.h>).
109  *      The process is stopped at the time read_regs is called.
110  *
111  * proc_write_regs(proc, regs)
112  *      Update the current register set from the passed in regs
113  *      structure.  Take care to avoid clobbering special CPU
114  *      registers or privileged bits in the PSL.
115  *      Depending on the architecture this may have fix-up work to do,
116  *      especially if the IAR or PCW are modified.
117  *      The process is stopped at the time write_regs is called.
118  *
119  * proc_read_fpregs, proc_write_fpregs
120  *      deal with the floating point register set, otherwise as above.
121  *
122  * proc_read_dbregs, proc_write_dbregs
123  *      deal with the processor debug register set, otherwise as above.
124  *
125  * proc_sstep(proc)
126  *      Arrange for the process to trap after executing a single instruction.
127  */
128
129 #define PROC_ACTION(action) do {                                        \
130         int error;                                                      \
131                                                                         \
132         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
133         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
134                 error = EIO;                                            \
135         else                                                            \
136                 error = (action);                                       \
137         return (error);                                                 \
138 } while(0)
139
140 int
141 proc_read_regs(struct thread *td, struct reg *regs)
142 {
143
144         PROC_ACTION(fill_regs(td, regs));
145 }
146
147 int
148 proc_write_regs(struct thread *td, struct reg *regs)
149 {
150
151         PROC_ACTION(set_regs(td, regs));
152 }
153
154 int
155 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
156 {
157
158         PROC_ACTION(fill_dbregs(td, dbregs));
159 }
160
161 int
162 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
163 {
164
165         PROC_ACTION(set_dbregs(td, dbregs));
166 }
167
168 /*
169  * Ptrace doesn't support fpregs at all, and there are no security holes
170  * or translations for fpregs, so we can just copy them.
171  */
172 int
173 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
174 {
175
176         PROC_ACTION(fill_fpregs(td, fpregs));
177 }
178
179 int
180 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
181 {
182
183         PROC_ACTION(set_fpregs(td, fpregs));
184 }
185
186 #ifdef COMPAT_FREEBSD32
187 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
188 int
189 proc_read_regs32(struct thread *td, struct reg32 *regs32)
190 {
191
192         PROC_ACTION(fill_regs32(td, regs32));
193 }
194
195 int
196 proc_write_regs32(struct thread *td, struct reg32 *regs32)
197 {
198
199         PROC_ACTION(set_regs32(td, regs32));
200 }
201
202 int
203 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
204 {
205
206         PROC_ACTION(fill_dbregs32(td, dbregs32));
207 }
208
209 int
210 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
211 {
212
213         PROC_ACTION(set_dbregs32(td, dbregs32));
214 }
215
216 int
217 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
218 {
219
220         PROC_ACTION(fill_fpregs32(td, fpregs32));
221 }
222
223 int
224 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
225 {
226
227         PROC_ACTION(set_fpregs32(td, fpregs32));
228 }
229 #endif
230
231 int
232 proc_sstep(struct thread *td)
233 {
234
235         PROC_ACTION(ptrace_single_step(td));
236 }
237
238 int
239 proc_rwmem(struct proc *p, struct uio *uio)
240 {
241         vm_map_t map;
242         vm_offset_t pageno;             /* page number */
243         vm_prot_t reqprot;
244         int error, fault_flags, page_offset, writing;
245
246         /*
247          * Assert that someone has locked this vmspace.  (Should be
248          * curthread but we can't assert that.)  This keeps the process
249          * from exiting out from under us until this operation completes.
250          */
251         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
252             p, p->p_pid));
253
254         /*
255          * The map we want...
256          */
257         map = &p->p_vmspace->vm_map;
258
259         /*
260          * If we are writing, then we request vm_fault() to create a private
261          * copy of each page.  Since these copies will not be writeable by the
262          * process, we must explicity request that they be dirtied.
263          */
264         writing = uio->uio_rw == UIO_WRITE;
265         reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
266         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
267
268         /*
269          * Only map in one page at a time.  We don't have to, but it
270          * makes things easier.  This way is trivial - right?
271          */
272         do {
273                 vm_offset_t uva;
274                 u_int len;
275                 vm_page_t m;
276
277                 uva = (vm_offset_t)uio->uio_offset;
278
279                 /*
280                  * Get the page number of this segment.
281                  */
282                 pageno = trunc_page(uva);
283                 page_offset = uva - pageno;
284
285                 /*
286                  * How many bytes to copy
287                  */
288                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
289
290                 /*
291                  * Fault and hold the page on behalf of the process.
292                  */
293                 error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
294                 if (error != KERN_SUCCESS) {
295                         if (error == KERN_RESOURCE_SHORTAGE)
296                                 error = ENOMEM;
297                         else
298                                 error = EFAULT;
299                         break;
300                 }
301
302                 /*
303                  * Now do the i/o move.
304                  */
305                 error = uiomove_fromphys(&m, page_offset, len, uio);
306
307                 /* Make the I-cache coherent for breakpoints. */
308                 if (writing && error == 0) {
309                         vm_map_lock_read(map);
310                         if (vm_map_check_protection(map, pageno, pageno +
311                             PAGE_SIZE, VM_PROT_EXECUTE))
312                                 vm_sync_icache(map, uva, len);
313                         vm_map_unlock_read(map);
314                 }
315
316                 /*
317                  * Release the page.
318                  */
319                 vm_page_lock(m);
320                 vm_page_unhold(m);
321                 vm_page_unlock(m);
322
323         } while (error == 0 && uio->uio_resid > 0);
324
325         return (error);
326 }
327
328 static int
329 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
330 {
331         struct vattr vattr;
332         vm_map_t map;
333         vm_map_entry_t entry;
334         vm_object_t obj, tobj, lobj;
335         struct vmspace *vm;
336         struct vnode *vp;
337         char *freepath, *fullpath;
338         u_int pathlen;
339         int error, index;
340
341         error = 0;
342         obj = NULL;
343
344         vm = vmspace_acquire_ref(p);
345         map = &vm->vm_map;
346         vm_map_lock_read(map);
347
348         do {
349                 entry = map->header.next;
350                 index = 0;
351                 while (index < pve->pve_entry && entry != &map->header) {
352                         entry = entry->next;
353                         index++;
354                 }
355                 if (index != pve->pve_entry) {
356                         error = EINVAL;
357                         break;
358                 }
359                 while (entry != &map->header &&
360                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
361                         entry = entry->next;
362                         index++;
363                 }
364                 if (entry == &map->header) {
365                         error = ENOENT;
366                         break;
367                 }
368
369                 /* We got an entry. */
370                 pve->pve_entry = index + 1;
371                 pve->pve_timestamp = map->timestamp;
372                 pve->pve_start = entry->start;
373                 pve->pve_end = entry->end - 1;
374                 pve->pve_offset = entry->offset;
375                 pve->pve_prot = entry->protection;
376
377                 /* Backing object's path needed? */
378                 if (pve->pve_pathlen == 0)
379                         break;
380
381                 pathlen = pve->pve_pathlen;
382                 pve->pve_pathlen = 0;
383
384                 obj = entry->object.vm_object;
385                 if (obj != NULL)
386                         VM_OBJECT_RLOCK(obj);
387         } while (0);
388
389         vm_map_unlock_read(map);
390         vmspace_free(vm);
391
392         pve->pve_fsid = VNOVAL;
393         pve->pve_fileid = VNOVAL;
394
395         if (error == 0 && obj != NULL) {
396                 lobj = obj;
397                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
398                         if (tobj != obj)
399                                 VM_OBJECT_RLOCK(tobj);
400                         if (lobj != obj)
401                                 VM_OBJECT_RUNLOCK(lobj);
402                         lobj = tobj;
403                         pve->pve_offset += tobj->backing_object_offset;
404                 }
405                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
406                 if (vp != NULL)
407                         vref(vp);
408                 if (lobj != obj)
409                         VM_OBJECT_RUNLOCK(lobj);
410                 VM_OBJECT_RUNLOCK(obj);
411
412                 if (vp != NULL) {
413                         freepath = NULL;
414                         fullpath = NULL;
415                         vn_fullpath(td, vp, &fullpath, &freepath);
416                         vn_lock(vp, LK_SHARED | LK_RETRY);
417                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
418                                 pve->pve_fileid = vattr.va_fileid;
419                                 pve->pve_fsid = vattr.va_fsid;
420                         }
421                         vput(vp);
422
423                         if (fullpath != NULL) {
424                                 pve->pve_pathlen = strlen(fullpath) + 1;
425                                 if (pve->pve_pathlen <= pathlen) {
426                                         error = copyout(fullpath, pve->pve_path,
427                                             pve->pve_pathlen);
428                                 } else
429                                         error = ENAMETOOLONG;
430                         }
431                         if (freepath != NULL)
432                                 free(freepath, M_TEMP);
433                 }
434         }
435         if (error == 0)
436                 CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
437                     p->p_pid, pve->pve_entry, pve->pve_start);
438
439         return (error);
440 }
441
442 #ifdef COMPAT_FREEBSD32
443 static int      
444 ptrace_vm_entry32(struct thread *td, struct proc *p,
445     struct ptrace_vm_entry32 *pve32)
446 {
447         struct ptrace_vm_entry pve;
448         int error;
449
450         pve.pve_entry = pve32->pve_entry;
451         pve.pve_pathlen = pve32->pve_pathlen;
452         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
453
454         error = ptrace_vm_entry(td, p, &pve);
455         if (error == 0) {
456                 pve32->pve_entry = pve.pve_entry;
457                 pve32->pve_timestamp = pve.pve_timestamp;
458                 pve32->pve_start = pve.pve_start;
459                 pve32->pve_end = pve.pve_end;
460                 pve32->pve_offset = pve.pve_offset;
461                 pve32->pve_prot = pve.pve_prot;
462                 pve32->pve_fileid = pve.pve_fileid;
463                 pve32->pve_fsid = pve.pve_fsid;
464         }
465
466         pve32->pve_pathlen = pve.pve_pathlen;
467         return (error);
468 }
469
470 static void
471 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
472     struct ptrace_lwpinfo32 *pl32)
473 {
474
475         pl32->pl_lwpid = pl->pl_lwpid;
476         pl32->pl_event = pl->pl_event;
477         pl32->pl_flags = pl->pl_flags;
478         pl32->pl_sigmask = pl->pl_sigmask;
479         pl32->pl_siglist = pl->pl_siglist;
480         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
481         strcpy(pl32->pl_tdname, pl->pl_tdname);
482         pl32->pl_child_pid = pl->pl_child_pid;
483 }
484 #endif /* COMPAT_FREEBSD32 */
485
486 /*
487  * Process debugging system call.
488  */
489 #ifndef _SYS_SYSPROTO_H_
490 struct ptrace_args {
491         int     req;
492         pid_t   pid;
493         caddr_t addr;
494         int     data;
495 };
496 #endif
497
498 #ifdef COMPAT_FREEBSD32
499 /*
500  * This CPP subterfuge is to try and reduce the number of ifdefs in
501  * the body of the code.
502  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
503  * becomes either:
504  *   copyin(uap->addr, &r.reg, sizeof r.reg);
505  * or
506  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
507  * .. except this is done at runtime.
508  */
509 #define COPYIN(u, k, s)         wrap32 ? \
510         copyin(u, k ## 32, s ## 32) : \
511         copyin(u, k, s)
512 #define COPYOUT(k, u, s)        wrap32 ? \
513         copyout(k ## 32, u, s ## 32) : \
514         copyout(k, u, s)
515 #else
516 #define COPYIN(u, k, s)         copyin(u, k, s)
517 #define COPYOUT(k, u, s)        copyout(k, u, s)
518 #endif
519 int
520 sys_ptrace(struct thread *td, struct ptrace_args *uap)
521 {
522         /*
523          * XXX this obfuscation is to reduce stack usage, but the register
524          * structs may be too large to put on the stack anyway.
525          */
526         union {
527                 struct ptrace_io_desc piod;
528                 struct ptrace_lwpinfo pl;
529                 struct ptrace_vm_entry pve;
530                 struct dbreg dbreg;
531                 struct fpreg fpreg;
532                 struct reg reg;
533 #ifdef COMPAT_FREEBSD32
534                 struct dbreg32 dbreg32;
535                 struct fpreg32 fpreg32;
536                 struct reg32 reg32;
537                 struct ptrace_io_desc32 piod32;
538                 struct ptrace_lwpinfo32 pl32;
539                 struct ptrace_vm_entry32 pve32;
540 #endif
541         } r;
542         void *addr;
543         int error = 0;
544 #ifdef COMPAT_FREEBSD32
545         int wrap32 = 0;
546
547         if (SV_CURPROC_FLAG(SV_ILP32))
548                 wrap32 = 1;
549 #endif
550         AUDIT_ARG_PID(uap->pid);
551         AUDIT_ARG_CMD(uap->req);
552         AUDIT_ARG_VALUE(uap->data);
553         addr = &r;
554         switch (uap->req) {
555         case PT_GETREGS:
556         case PT_GETFPREGS:
557         case PT_GETDBREGS:
558         case PT_LWPINFO:
559                 break;
560         case PT_SETREGS:
561                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
562                 break;
563         case PT_SETFPREGS:
564                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
565                 break;
566         case PT_SETDBREGS:
567                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
568                 break;
569         case PT_IO:
570                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
571                 break;
572         case PT_VM_ENTRY:
573                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
574                 break;
575         default:
576                 addr = uap->addr;
577                 break;
578         }
579         if (error)
580                 return (error);
581
582         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
583         if (error)
584                 return (error);
585
586         switch (uap->req) {
587         case PT_VM_ENTRY:
588                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
589                 break;
590         case PT_IO:
591                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
592                 break;
593         case PT_GETREGS:
594                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
595                 break;
596         case PT_GETFPREGS:
597                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
598                 break;
599         case PT_GETDBREGS:
600                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
601                 break;
602         case PT_LWPINFO:
603                 error = copyout(&r.pl, uap->addr, uap->data);
604                 break;
605         }
606
607         return (error);
608 }
609 #undef COPYIN
610 #undef COPYOUT
611
612 #ifdef COMPAT_FREEBSD32
613 /*
614  *   PROC_READ(regs, td2, addr);
615  * becomes either:
616  *   proc_read_regs(td2, addr);
617  * or
618  *   proc_read_regs32(td2, addr);
619  * .. except this is done at runtime.  There is an additional
620  * complication in that PROC_WRITE disallows 32 bit consumers
621  * from writing to 64 bit address space targets.
622  */
623 #define PROC_READ(w, t, a)      wrap32 ? \
624         proc_read_ ## w ## 32(t, a) : \
625         proc_read_ ## w (t, a)
626 #define PROC_WRITE(w, t, a)     wrap32 ? \
627         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
628         proc_write_ ## w (t, a)
629 #else
630 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
631 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
632 #endif
633
634 int
635 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
636 {
637         struct iovec iov;
638         struct uio uio;
639         struct proc *curp, *p, *pp;
640         struct thread *td2 = NULL, *td3;
641         struct ptrace_io_desc *piod = NULL;
642         struct ptrace_lwpinfo *pl;
643         int error, write, tmp, num;
644         int proctree_locked = 0;
645         lwpid_t tid = 0, *buf;
646 #ifdef COMPAT_FREEBSD32
647         int wrap32 = 0, safe = 0;
648         struct ptrace_io_desc32 *piod32 = NULL;
649         struct ptrace_lwpinfo32 *pl32 = NULL;
650         struct ptrace_lwpinfo plr;
651 #endif
652
653         curp = td->td_proc;
654
655         /* Lock proctree before locking the process. */
656         switch (req) {
657         case PT_TRACE_ME:
658         case PT_ATTACH:
659         case PT_STEP:
660         case PT_CONTINUE:
661         case PT_TO_SCE:
662         case PT_TO_SCX:
663         case PT_SYSCALL:
664         case PT_FOLLOW_FORK:
665         case PT_DETACH:
666                 sx_xlock(&proctree_lock);
667                 proctree_locked = 1;
668                 break;
669         default:
670                 break;
671         }
672
673         write = 0;
674         if (req == PT_TRACE_ME) {
675                 p = td->td_proc;
676                 PROC_LOCK(p);
677         } else {
678                 if (pid <= PID_MAX) {
679                         if ((p = pfind(pid)) == NULL) {
680                                 if (proctree_locked)
681                                         sx_xunlock(&proctree_lock);
682                                 return (ESRCH);
683                         }
684                 } else {
685                         td2 = tdfind(pid, -1);
686                         if (td2 == NULL) {
687                                 if (proctree_locked)
688                                         sx_xunlock(&proctree_lock);
689                                 return (ESRCH);
690                         }
691                         p = td2->td_proc;
692                         tid = pid;
693                         pid = p->p_pid;
694                 }
695         }
696         AUDIT_ARG_PROCESS(p);
697
698         if ((p->p_flag & P_WEXIT) != 0) {
699                 error = ESRCH;
700                 goto fail;
701         }
702         if ((error = p_cansee(td, p)) != 0)
703                 goto fail;
704
705         if ((error = p_candebug(td, p)) != 0)
706                 goto fail;
707
708         /*
709          * System processes can't be debugged.
710          */
711         if ((p->p_flag & P_SYSTEM) != 0) {
712                 error = EINVAL;
713                 goto fail;
714         }
715
716         if (tid == 0) {
717                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
718                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
719                         td2 = p->p_xthread;
720                 } else {
721                         td2 = FIRST_THREAD_IN_PROC(p);
722                 }
723                 tid = td2->td_tid;
724         }
725
726 #ifdef COMPAT_FREEBSD32
727         /*
728          * Test if we're a 32 bit client and what the target is.
729          * Set the wrap controls accordingly.
730          */
731         if (SV_CURPROC_FLAG(SV_ILP32)) {
732                 if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
733                         safe = 1;
734                 wrap32 = 1;
735         }
736 #endif
737         /*
738          * Permissions check
739          */
740         switch (req) {
741         case PT_TRACE_ME:
742                 /* Always legal. */
743                 break;
744
745         case PT_ATTACH:
746                 /* Self */
747                 if (p->p_pid == td->td_proc->p_pid) {
748                         error = EINVAL;
749                         goto fail;
750                 }
751
752                 /* Already traced */
753                 if (p->p_flag & P_TRACED) {
754                         error = EBUSY;
755                         goto fail;
756                 }
757
758                 /* Can't trace an ancestor if you're being traced. */
759                 if (curp->p_flag & P_TRACED) {
760                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
761                                 if (pp == p) {
762                                         error = EINVAL;
763                                         goto fail;
764                                 }
765                         }
766                 }
767
768
769                 /* OK */
770                 break;
771
772         case PT_CLEARSTEP:
773                 /* Allow thread to clear single step for itself */
774                 if (td->td_tid == tid)
775                         break;
776
777                 /* FALLTHROUGH */
778         default:
779                 /* not being traced... */
780                 if ((p->p_flag & P_TRACED) == 0) {
781                         error = EPERM;
782                         goto fail;
783                 }
784
785                 /* not being traced by YOU */
786                 if (p->p_pptr != td->td_proc) {
787                         error = EBUSY;
788                         goto fail;
789                 }
790
791                 /* not currently stopped */
792                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
793                     p->p_suspcount != p->p_numthreads  ||
794                     (p->p_flag & P_WAITED) == 0) {
795                         error = EBUSY;
796                         goto fail;
797                 }
798
799                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
800                         static int count = 0;
801                         if (count++ == 0)
802                                 printf("P_STOPPED_TRACE not set.\n");
803                 }
804
805                 /* OK */
806                 break;
807         }
808
809         /* Keep this process around until we finish this request. */
810         _PHOLD(p);
811
812 #ifdef FIX_SSTEP
813         /*
814          * Single step fixup ala procfs
815          */
816         FIX_SSTEP(td2);
817 #endif
818
819         /*
820          * Actually do the requests
821          */
822
823         td->td_retval[0] = 0;
824
825         switch (req) {
826         case PT_TRACE_ME:
827                 /* set my trace flag and "owner" so it can read/write me */
828                 p->p_flag |= P_TRACED;
829                 if (p->p_flag & P_PPWAIT)
830                         p->p_flag |= P_PPTRACE;
831                 p->p_oppid = p->p_pptr->p_pid;
832                 CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
833                 break;
834
835         case PT_ATTACH:
836                 /* security check done above */
837                 /*
838                  * It would be nice if the tracing relationship was separate
839                  * from the parent relationship but that would require
840                  * another set of links in the proc struct or for "wait"
841                  * to scan the entire proc table.  To make life easier,
842                  * we just re-parent the process we're trying to trace.
843                  * The old parent is remembered so we can put things back
844                  * on a "detach".
845                  */
846                 p->p_flag |= P_TRACED;
847                 p->p_oppid = p->p_pptr->p_pid;
848                 if (p->p_pptr != td->td_proc) {
849                         proc_reparent(p, td->td_proc);
850                 }
851                 data = SIGSTOP;
852                 CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
853                     p->p_oppid);
854                 goto sendsig;   /* in PT_CONTINUE below */
855
856         case PT_CLEARSTEP:
857                 CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
858                     p->p_pid);
859                 error = ptrace_clear_single_step(td2);
860                 break;
861
862         case PT_SETSTEP:
863                 CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
864                     p->p_pid);
865                 error = ptrace_single_step(td2);
866                 break;
867
868         case PT_SUSPEND:
869                 CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
870                     p->p_pid);
871                 td2->td_dbgflags |= TDB_SUSPEND;
872                 thread_lock(td2);
873                 td2->td_flags |= TDF_NEEDSUSPCHK;
874                 thread_unlock(td2);
875                 break;
876
877         case PT_RESUME:
878                 CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
879                     p->p_pid);
880                 td2->td_dbgflags &= ~TDB_SUSPEND;
881                 break;
882
883         case PT_FOLLOW_FORK:
884                 CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
885                     p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
886                     data ? "enabled" : "disabled");
887                 if (data)
888                         p->p_flag |= P_FOLLOWFORK;
889                 else
890                         p->p_flag &= ~P_FOLLOWFORK;
891                 break;
892
893         case PT_STEP:
894         case PT_CONTINUE:
895         case PT_TO_SCE:
896         case PT_TO_SCX:
897         case PT_SYSCALL:
898         case PT_DETACH:
899                 /* Zero means do not send any signal */
900                 if (data < 0 || data > _SIG_MAXSIG) {
901                         error = EINVAL;
902                         break;
903                 }
904
905                 switch (req) {
906                 case PT_STEP:
907                         CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
908                             td2->td_tid, p->p_pid);
909                         error = ptrace_single_step(td2);
910                         if (error)
911                                 goto out;
912                         break;
913                 case PT_CONTINUE:
914                 case PT_TO_SCE:
915                 case PT_TO_SCX:
916                 case PT_SYSCALL:
917                         if (addr != (void *)1) {
918                                 error = ptrace_set_pc(td2,
919                                     (u_long)(uintfptr_t)addr);
920                                 if (error)
921                                         goto out;
922                         }
923                         switch (req) {
924                         case PT_TO_SCE:
925                                 p->p_stops |= S_PT_SCE;
926                                 CTR2(KTR_PTRACE,
927                                     "PT_TO_SCE: pid %d, stops = %#x", p->p_pid,
928                                     p->p_stops);
929                                 break;
930                         case PT_TO_SCX:
931                                 p->p_stops |= S_PT_SCX;
932                                 CTR2(KTR_PTRACE,
933                                     "PT_TO_SCX: pid %d, stops = %#x", p->p_pid,
934                                     p->p_stops);
935                                 break;
936                         case PT_SYSCALL:
937                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
938                                 CTR2(KTR_PTRACE,
939                                     "PT_SYSCALL: pid %d, stops = %#x", p->p_pid,
940                                     p->p_stops);
941                                 break;
942                         case PT_CONTINUE:
943                                 CTR1(KTR_PTRACE,
944                                     "PT_CONTINUE: pid %d", p->p_pid);
945                                 break;
946                         }
947                         break;
948                 case PT_DETACH:
949                         /* reset process parent */
950                         if (p->p_oppid != p->p_pptr->p_pid) {
951                                 PROC_LOCK(p->p_pptr);
952                                 sigqueue_take(p->p_ksi);
953                                 PROC_UNLOCK(p->p_pptr);
954
955                                 pp = proc_realparent(p);
956                                 proc_reparent(p, pp);
957                                 if (pp == initproc)
958                                         p->p_sigparent = SIGCHLD;
959                                 CTR2(KTR_PTRACE,
960                                     "PT_DETACH: pid %d reparented to pid %d",
961                                     p->p_pid, pp->p_pid);
962                         } else
963                                 CTR1(KTR_PTRACE, "PT_DETACH: pid %d", p->p_pid);
964                         p->p_oppid = 0;
965                         p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
966                         p->p_stops = 0;
967
968                         /* should we send SIGCHLD? */
969                         /* childproc_continued(p); */
970                         break;
971                 }
972
973         sendsig:
974                 if (proctree_locked) {
975                         sx_xunlock(&proctree_lock);
976                         proctree_locked = 0;
977                 }
978                 p->p_xstat = data;
979                 p->p_xthread = NULL;
980                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
981                         /* deliver or queue signal */
982                         td2->td_dbgflags &= ~TDB_XSIG;
983                         td2->td_xsig = data;
984
985                         if (req == PT_DETACH) {
986                                 FOREACH_THREAD_IN_PROC(p, td3)
987                                         td3->td_dbgflags &= ~TDB_SUSPEND; 
988                         }
989                         /*
990                          * unsuspend all threads, to not let a thread run,
991                          * you should use PT_SUSPEND to suspend it before
992                          * continuing process.
993                          */
994                         PROC_SLOCK(p);
995                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
996                         thread_unsuspend(p);
997                         PROC_SUNLOCK(p);
998                         if (req == PT_ATTACH)
999                                 kern_psignal(p, data);
1000                 } else {
1001                         if (data)
1002                                 kern_psignal(p, data);
1003                 }
1004                 break;
1005
1006         case PT_WRITE_I:
1007         case PT_WRITE_D:
1008                 td2->td_dbgflags |= TDB_USERWR;
1009                 write = 1;
1010                 /* FALLTHROUGH */
1011         case PT_READ_I:
1012         case PT_READ_D:
1013                 PROC_UNLOCK(p);
1014                 tmp = 0;
1015                 /* write = 0 set above */
1016                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1017                 iov.iov_len = sizeof(int);
1018                 uio.uio_iov = &iov;
1019                 uio.uio_iovcnt = 1;
1020                 uio.uio_offset = (off_t)(uintptr_t)addr;
1021                 uio.uio_resid = sizeof(int);
1022                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1023                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1024                 uio.uio_td = td;
1025                 error = proc_rwmem(p, &uio);
1026                 if (uio.uio_resid != 0) {
1027                         /*
1028                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1029                          * so I think write() can bogusly return 0.
1030                          * XXX what happens for short writes?  We don't want
1031                          * to write partial data.
1032                          * XXX proc_rwmem() returns EPERM for other invalid
1033                          * addresses.  Convert this to EINVAL.  Does this
1034                          * clobber returns of EPERM for other reasons?
1035                          */
1036                         if (error == 0 || error == ENOSPC || error == EPERM)
1037                                 error = EINVAL; /* EOF */
1038                 }
1039                 if (!write)
1040                         td->td_retval[0] = tmp;
1041                 if (error == 0) {
1042                         if (write)
1043                                 CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1044                                     p->p_pid, addr, data);
1045                         else
1046                                 CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1047                                     p->p_pid, addr, tmp);
1048                 }
1049                 PROC_LOCK(p);
1050                 break;
1051
1052         case PT_IO:
1053 #ifdef COMPAT_FREEBSD32
1054                 if (wrap32) {
1055                         piod32 = addr;
1056                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1057                         iov.iov_len = piod32->piod_len;
1058                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1059                         uio.uio_resid = piod32->piod_len;
1060                 } else
1061 #endif
1062                 {
1063                         piod = addr;
1064                         iov.iov_base = piod->piod_addr;
1065                         iov.iov_len = piod->piod_len;
1066                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1067                         uio.uio_resid = piod->piod_len;
1068                 }
1069                 uio.uio_iov = &iov;
1070                 uio.uio_iovcnt = 1;
1071                 uio.uio_segflg = UIO_USERSPACE;
1072                 uio.uio_td = td;
1073 #ifdef COMPAT_FREEBSD32
1074                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1075 #else
1076                 tmp = piod->piod_op;
1077 #endif
1078                 switch (tmp) {
1079                 case PIOD_READ_D:
1080                 case PIOD_READ_I:
1081                         CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1082                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1083                         uio.uio_rw = UIO_READ;
1084                         break;
1085                 case PIOD_WRITE_D:
1086                 case PIOD_WRITE_I:
1087                         CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1088                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1089                         td2->td_dbgflags |= TDB_USERWR;
1090                         uio.uio_rw = UIO_WRITE;
1091                         break;
1092                 default:
1093                         error = EINVAL;
1094                         goto out;
1095                 }
1096                 PROC_UNLOCK(p);
1097                 error = proc_rwmem(p, &uio);
1098 #ifdef COMPAT_FREEBSD32
1099                 if (wrap32)
1100                         piod32->piod_len -= uio.uio_resid;
1101                 else
1102 #endif
1103                         piod->piod_len -= uio.uio_resid;
1104                 PROC_LOCK(p);
1105                 break;
1106
1107         case PT_KILL:
1108                 CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1109                 data = SIGKILL;
1110                 goto sendsig;   /* in PT_CONTINUE above */
1111
1112         case PT_SETREGS:
1113                 CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1114                     p->p_pid);
1115                 td2->td_dbgflags |= TDB_USERWR;
1116                 error = PROC_WRITE(regs, td2, addr);
1117                 break;
1118
1119         case PT_GETREGS:
1120                 CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1121                     p->p_pid);
1122                 error = PROC_READ(regs, td2, addr);
1123                 break;
1124
1125         case PT_SETFPREGS:
1126                 CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1127                     p->p_pid);
1128                 td2->td_dbgflags |= TDB_USERWR;
1129                 error = PROC_WRITE(fpregs, td2, addr);
1130                 break;
1131
1132         case PT_GETFPREGS:
1133                 CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1134                     p->p_pid);
1135                 error = PROC_READ(fpregs, td2, addr);
1136                 break;
1137
1138         case PT_SETDBREGS:
1139                 CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1140                     p->p_pid);
1141                 td2->td_dbgflags |= TDB_USERWR;
1142                 error = PROC_WRITE(dbregs, td2, addr);
1143                 break;
1144
1145         case PT_GETDBREGS:
1146                 CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1147                     p->p_pid);
1148                 error = PROC_READ(dbregs, td2, addr);
1149                 break;
1150
1151         case PT_LWPINFO:
1152                 if (data <= 0 ||
1153 #ifdef COMPAT_FREEBSD32
1154                     (!wrap32 && data > sizeof(*pl)) ||
1155                     (wrap32 && data > sizeof(*pl32))) {
1156 #else
1157                     data > sizeof(*pl)) {
1158 #endif
1159                         error = EINVAL;
1160                         break;
1161                 }
1162 #ifdef COMPAT_FREEBSD32
1163                 if (wrap32) {
1164                         pl = &plr;
1165                         pl32 = addr;
1166                 } else
1167 #endif
1168                 pl = addr;
1169                 pl->pl_lwpid = td2->td_tid;
1170                 pl->pl_event = PL_EVENT_NONE;
1171                 pl->pl_flags = 0;
1172                 if (td2->td_dbgflags & TDB_XSIG) {
1173                         pl->pl_event = PL_EVENT_SIGNAL;
1174                         if (td2->td_dbgksi.ksi_signo != 0 &&
1175 #ifdef COMPAT_FREEBSD32
1176                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1177                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1178                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1179                             pl_siginfo) + sizeof(struct siginfo32)))
1180 #else
1181                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1182                             + sizeof(pl->pl_siginfo)
1183 #endif
1184                         ){
1185                                 pl->pl_flags |= PL_FLAG_SI;
1186                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1187                         }
1188                 }
1189                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1190                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1191                 if (td2->td_dbgflags & TDB_SCE)
1192                         pl->pl_flags |= PL_FLAG_SCE;
1193                 else if (td2->td_dbgflags & TDB_SCX)
1194                         pl->pl_flags |= PL_FLAG_SCX;
1195                 if (td2->td_dbgflags & TDB_EXEC)
1196                         pl->pl_flags |= PL_FLAG_EXEC;
1197                 if (td2->td_dbgflags & TDB_FORK) {
1198                         pl->pl_flags |= PL_FLAG_FORKED;
1199                         pl->pl_child_pid = td2->td_dbg_forked;
1200                 }
1201                 if (td2->td_dbgflags & TDB_CHILD)
1202                         pl->pl_flags |= PL_FLAG_CHILD;
1203                 pl->pl_sigmask = td2->td_sigmask;
1204                 pl->pl_siglist = td2->td_siglist;
1205                 strcpy(pl->pl_tdname, td2->td_name);
1206 #ifdef COMPAT_FREEBSD32
1207                 if (wrap32)
1208                         ptrace_lwpinfo_to32(pl, pl32);
1209 #endif
1210                 CTR5(KTR_PTRACE,
1211             "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d",
1212                     td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1213                     pl->pl_child_pid);
1214                 break;
1215
1216         case PT_GETNUMLWPS:
1217                 CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1218                     p->p_numthreads);
1219                 td->td_retval[0] = p->p_numthreads;
1220                 break;
1221
1222         case PT_GETLWPLIST:
1223                 CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1224                     p->p_pid, data, p->p_numthreads);
1225                 if (data <= 0) {
1226                         error = EINVAL;
1227                         break;
1228                 }
1229                 num = imin(p->p_numthreads, data);
1230                 PROC_UNLOCK(p);
1231                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1232                 tmp = 0;
1233                 PROC_LOCK(p);
1234                 FOREACH_THREAD_IN_PROC(p, td2) {
1235                         if (tmp >= num)
1236                                 break;
1237                         buf[tmp++] = td2->td_tid;
1238                 }
1239                 PROC_UNLOCK(p);
1240                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1241                 free(buf, M_TEMP);
1242                 if (!error)
1243                         td->td_retval[0] = tmp;
1244                 PROC_LOCK(p);
1245                 break;
1246
1247         case PT_VM_TIMESTAMP:
1248                 CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1249                     p->p_pid, p->p_vmspace->vm_map.timestamp);
1250                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1251                 break;
1252
1253         case PT_VM_ENTRY:
1254                 PROC_UNLOCK(p);
1255 #ifdef COMPAT_FREEBSD32
1256                 if (wrap32)
1257                         error = ptrace_vm_entry32(td, p, addr);
1258                 else
1259 #endif
1260                 error = ptrace_vm_entry(td, p, addr);
1261                 PROC_LOCK(p);
1262                 break;
1263
1264         default:
1265 #ifdef __HAVE_PTRACE_MACHDEP
1266                 if (req >= PT_FIRSTMACH) {
1267                         PROC_UNLOCK(p);
1268                         error = cpu_ptrace(td2, req, addr, data);
1269                         PROC_LOCK(p);
1270                 } else
1271 #endif
1272                         /* Unknown request. */
1273                         error = EINVAL;
1274                 break;
1275         }
1276
1277 out:
1278         /* Drop our hold on this process now that the request has completed. */
1279         _PRELE(p);
1280 fail:
1281         PROC_UNLOCK(p);
1282         if (proctree_locked)
1283                 sx_xunlock(&proctree_lock);
1284         return (error);
1285 }
1286 #undef PROC_READ
1287 #undef PROC_WRITE
1288
1289 /*
1290  * Stop a process because of a debugging event;
1291  * stay stopped until p->p_step is cleared
1292  * (cleared by PIOCCONT in procfs).
1293  */
1294 void
1295 stopevent(struct proc *p, unsigned int event, unsigned int val)
1296 {
1297
1298         PROC_LOCK_ASSERT(p, MA_OWNED);
1299         p->p_step = 1;
1300         CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1301             val);
1302         do {
1303                 p->p_xstat = val;
1304                 p->p_xthread = NULL;
1305                 p->p_stype = event;     /* Which event caused the stop? */
1306                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1307                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1308         } while (p->p_step);
1309 }